Deep Learning for Medical Imaging

6 EC

Semester 2, period 5

5354DLFM6Y

Owner Master Physics and Astronomy (joint degree)
Coordinator dr. O.J. Gurney-Champion
Part of Master Physics and Astronomy (joint degree), track Biophysics and Biophotonics, Master Physics and Astronomy (joint degree), track General Physics and Astronomy,

Course manual 2022/2023

Course content

Artificial Intelligence has proven to be a great tool for helping radiologists and pathologists in diagnosing patients, and, ultimately, selecting the best possible patient-specific treatment. Computers can analyse digital images at an unmet speed and can detect patterns that are missed by medical experts. The development and application of artificial intelligence in medical imaging has sped up due to (1) widely available digital medical images, (2) freely available machine learning tools, and (3) high computing performance and GPU’s in particular. This combination has led to applications where computers are highly accurate in detecting patterns, and lesions to support diagnosis and prognosis. AI is used over the entire front of medical imaging, including designing optimal image acquisition schemes, acceleration of imaging, reconstruction of imaging, image enhancement, segmentation and classification.

In this course, we will focus on applying deep learning for digital medical image acquisition, processing and automatic analysis of images. The course will introduce the basic concepts of deep learning. Students will get hands-on experience in using the most common deep neural networks that are used in medical imaging, including convolutional neural networks such as U-net. Ultimately, the core of the course will focus on combining your technical background and knowledge about physics to allow you to apply and enhance deep learning approaches for medical imaging.

Study materials

Literature

  • Understanding Deep Learning by Simon J.D. Prince: https://udlbook.github.io/udlbook/

Syllabus

  • Online on Canvas

Practical training material

  • 3 extensive exercises

Software

  • Python

  • Pytorch

  • Lisa and Collabnet

Objectives

  • The student can explain the principles and use of neural computing via feed-forward neural networks and, particularly, convolutional neural networks.
  • The student can calculate a forward pass and a back-propagation through a simple fully connected neural network
  • The student can apply deep learning architectures to medical images.
  • The student can analyse the performance of different deep learning architectures in the context of a deep learning challenge.
  • The student can use their insight in technical aspects of medical imaging to construct networks that solve specific technical aspects in the imaging pipeline.
  • The student can justify the use of different physics aspects in deep learning.

Teaching methods

  • Lecture
  • Computer lab session/practical training
  • Presentation/symposium
  • Self-study

This course contains some self-study via Canvas, such that the students come well-prepared to the classes.

The classes will be interactive and activating.

We have 3 blocks, each block contains 1 week of predominantly theory and 1-2 weeks of hands-on programming of neural networks.

The final exercise will be presented as poster to your peers. 

 

Learning activities

Activity

Hours

  

Classes

14

 (7 classes x 2 hours)
Practicals 21 (7 practicals x 3 hours)
Exercise class 9 (3 classes x 3 hours)
Self study 45  

Self programming

79

  

Total

168

(6 EC x 28 hours)

Attendance

Requirements concerning attendance (OER-B).

  • In addition to, or instead of, classes in the form of lectures, the elements of the master’s examination programme often include a practical component as defined in article A-1.2 of part A. The course catalogue contains information on the types of classes in each part of the programme. Attendance during practical components is mandatory.

    • Additional requirements for this course:

    Participation in the classes is expected and only 1 out of 7 classes can be missed.

    Contact the coördinator if you are missing a class.

    Although practicals are not mandatory, they are very essential for students to successfully hand in their exercises.

    Participants must prepare the classes via self study and the study done via Perusall will contribute for 20% to their marks. 

    Assessment

    Item and weight Details

    Final grade

    20%

    Persuall intro understanding deep learning book

    25%

    Block 1: my first network

    25%

    Block 2: CNN

    20%

    Block 3: Image reconstruction

    10%

    Poster pressentatie

    Assignments

    The course consists of 3 blocks, each with an assignment.

    Block 1: Introduction to neural networks

    Block 2: Convolutional Neural Networks

    Block 3: Deep learning for image reconstruction

    Each assignment can be done in pairs and counts towards your grade.

    The assignments will be graded by TAs and feedback will be provided.

    Furthermore, a Perusall assignment will be assigned to prepare for the lectures. Perusall will assign a grade to this assignment. The teachers will manually go through the comments and take meaningful remarks into account in the Perusall assessment.

    At the end of the course, a mini-symposium will be held in which the students present the results of Block 3 to each other. The posters and discussion will be evaluated by the TAs and teachers.

    Fraud and plagiarism

    The 'Regulations governing fraud and plagiarism for UvA students' applies to this course. This will be monitored carefully. Upon suspicion of fraud or plagiarism the Examinations Board of the programme will be informed. For the 'Regulations governing fraud and plagiarism for UvA students' see: www.student.uva.nl

    Course structure

    Weeknummer Onderwerpen Studiestof Deadlines
    1  Intro to deep learning chapters 2-7 of book; canvas 2 x Perusall
    2 My first Numpy network chapter 8; canvas; exercises Perusall
    3 My first PyTorch network chapter 9; canvas; exercises Perusall; First exercise 
    4 Convolutional networks chapter 10; canvas Perusall
    5 Convolutional networks canvas Second exercise
    6 Image reconstruction canvas Perusall
    7 Image reconstruction canvas  
    8 finishing up   Third exercise; Poster presentation

    Timetable

    The schedule for this course is published on DataNose.

    Additional information

    The students should be familiar with the basics of programming in Python.

    The students should have a basic understanding of medical imaging.

    No prior experience in machine learning is expected/required.

    Contact information

    Coordinator

    • dr. O.J. Gurney-Champion

    Staff

    • Matthan Caan